This morning’s Gamelab-talk by PlayStation 4 architect Mark Cerny was a parade of revelations on the development history of the new console. One of the bits Cerny disclosed was that he was not asked to become the Lead System Architect. Cerny put himself foward and Sony accepted.
Cerny is hyper intelligent and he is not afraid to share that with his audience. He tells that at the age of 16 he is already accepted as a student at the University of Berkeley “Two years younger than the average.” It turns out he finds the study boring and a year later Cerny is hired by Atari. “I was the youngest employee by far, at least five years younger than the others. Atari thought it was cool to have a child prodigy programmer working for them.”
Cerny gets away with patting himself on the back because he tells the story in the same businesslike fashion as the rest of his presentation — he kind of resembles Sheldon from the Big Bang Theory. He talks about his adventures at SEGA, where he works on the first Sonic and becomes fluent in Japanese. Then about the time he spends at Crystal Dynamics as a Graphics Engine Developer and about his adventures at Universal Interactive Studios, where he gets promoted to President and scouts studios like Naughty Dog an Insomniac.
It is his friend Shuhei Yoshida, the current President of Sony Worldwide Studios, who convinces Cerny to leave Universal so he can get involved with the PlayStation 2 — the console that Sony is working on at that time. Cerny develops the first engines for the console and discovers the immense complexity of the Emotion Engine, the core of the PS2.
To describe the complexity he uses the term Time to Triangle, meaning the time a studio needs on a new console to get from scratch to the first working 3D levels. With the PlayStation 1 it took developers 1-2 months. With the PS2 this time has more than doubled to 3-6 months. Cerny launches the Initiative for a Common Engine, or in short ICE to counter the long development cycle and the ever growing teams.
The ICE team is based at Naughty Dog and it actually succeeds in making the necessary tools for PS2-development. When the development of the PS3 starts, the ICE team gets direct access to the hardware team. Within a couple of months, the ICE team, consisting of a couple of dozen people, masters the new Cellprocessor. They become over-confident which results in a disastrous attitude: ICE sees third parties as competition “We’re gonna show EA and Rockstar who has the best graphics engine!”
A couple of months before launch it gets apparent how stupid that attitude actually is. “Third party developers did not understand Cell at all and all of a sudden ICE realises that third parties are not the competition, but an important part of the platform. Competitors become collaborators. But it is already too late. This results in the poor launch line-up of the PlayStation 3.
Another problem is the Time to Triangle for the new console. This was already bad with the PS2, but with the PS3 this has increased to a staggering 6-12 months, according to Cerny. The moment he realizes this will turn out to be crucial for the development philosophy of the PlayStation 4
As early as 2007, Sony does a post mortem for the PS3 and looks ahead at the new generation. At that point, the hardware team has a strong preference for a new Cell successor. When Cerny suggests de x86-architecture the reply is: “That’s unusable in a console!” How can consoles compete with PC’s that continously grow stronger when both use the same architecture? PowerPC-cpu’s, like the Cell, are the better option. Harder to develop for, but ultimately more powerful.
Cerny is not convinced. He studies the thirty year history of the x86. During his vacation. Out of passion for the subject. That sparks an idea. Cerny rechecks his resume. He is American, but fluent in Japanese, he has been a producer, game designer (Ratchet & Clank) and an engine programmer. He was already involved in the development of the PlayStation 2 and 3. Mark Cerny realises that who other than Mark Cerny is the best qualified person to develop the PlayStation 4.
He suggests this idea to Yoshida, who likes it. And when Ken Katuragi also concurs, Cerny is indeed the new Lead System Architect for the PlayStation 4. This means that he has to let go of his position at Sony Worldwide Studios. That, however, is not a problem, since Cerny is not a real employee. He’s a freelance consultant working for Sony through his own company Cerny Games. And he still is. So in fact, Sony’s new console has not been designed by a Sony employee.
Cerny throws himself on his new task an visits dozens of studios and sends out questionnaires to anyone who even remotely considers to develop a game for a Sony console. The most important feedback he gets is surprising:
- Unified Memory
- A CPU-count of 4 or 8
- Invest in a powerful GPU
- “Not too exotic”
This partially reads like the anti-PS3. Based on this Cerny settles on his design motto, borrowed from Nolan Bushnell: Easy to Learn, Hard to Master.
Easy to Learn is according to Cerny an absolute must to decrease the Time to Triangle for PlayStation 4. That is the mindset that prevents the comeback of the Cell-processor. This is the reason why Sony ultimately choose the x86-architectuur.
But that’s not all. Also in the rest of the structure, like the memory bandwith, Easy to Learn prevails. Through a technical story, Cerny explains why they sacrificed a lot of raw speed.
The final design of the PS4 has GDDR5 chips and a 256 bit bus. The bandwidth is 176 Gigabyte/second. A lot. Sony also considered another option that would result in an impressive bandwidth of 1088 Gb/s. That option consisted of a 128 bit bus that in principle only would generate 88 Gb/s, but the implementation of small eDram would add a whopping 1000 Gb/s.
Using eDram, however, is quite complicated. I would take developers months to optimize their engines so that they would be able to actually utilise that speed. The opposite of Easy to Learn and therefore Cerny cancelled the eDram.
The Hard to Master part is translated by Cerny into a rich feature set. The PS4 is jam packed with advanced options that do take some time to be mastered. This results in a console that developers can start working with in short term, but that also offers opportunities to explore in the long run. Indeed Easy to Learn, Hard to Master.
And the Time to Triangle for the Ps4? According to Cerny only 1-2 months. Mission accomplished.